AI BriefingKokon 0.10 adds Qwen-Image 2.1 prompt generation and Windows-only LoRA support
·2026.10.21 03:16
Key point
Kokon 0.10 introduces a 'Prompt+' feature that uses the existing Qwen3-VL 8B encoder to expand prompts, adding roughly 2 seconds of processing time on an RTX 5070 Ti, alongside Windows-only support for custom Qwen-Image 2.1 LoRAs.
Details
- Kokon 0.10 introduces a "Prompt+" feature that expands short user ideas into detailed prompts using Qwen3-VL 8B, the text encoder already loaded by Qwen-Image 2.1, requiring no additional model download or memory overhead.
- The feature adds approximately 2 seconds per image on an RTX 5070 Ti and 13 seconds on an M2 Pro, and remains functional on 8 GB graphics cards.
- Users can now load custom Qwen-Image 2.1 LoRAs (beta) with adjustable strength and trigger words, supporting up to three LoRAs simultaneously; this feature is currently tested on Windows with NVIDIA only, and LoRAs for older versions (1.0, 2512, Edit 2509/2511) are rejected.
- The application now runs a custom build of stable-diffusion.cpp on Windows, macOS, and Linux, achieving about 127 tokens/s for text generation on a 5070 Ti while producing bit-identical images to the upstream code.
- The software is free and open source under the AGPL license, with weights subject to the Qwen Research License (non-commercial), and requires a one-time engine download of 29–352 MB depending on the platform.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
ArtCraft releases open-source IDE for AI image and video creation with 62-model catalog